AndrewGrant wrote: ↑Mon Jul 05, 2021 6:33 pm
Based on a quick few thousand games ... I can promise its a ponder issue. Why do I have pondering in this engine again? Time to rip it out smh.
It's weird. I changed the avx2 binary to avx, and it still crashed. But now it's playing against Vajolet and so far it looks fine (vs Chiron it crashes almost instantly). Maybe it's opponent-dependant?
I cannot understand why 12.75 works perfectly on Windows 10 & 11 ARM64 and 13.00 not
At infinite level 13:00 never crashes. When playing a game it starts thinking, I see the hash increasing eg: -> 36 MB... -> 177 MB -> 178 MB then it fail to 0 Mb and never makes a move, losing on time. If I set only 1 CPU, it plays well, crashing rarely. Must be something related to multithreading or hash memory. Windows 11 ARM64 will be on every Surface device on November. I have tried also all the newer builds on github.
Since I like Ethereal since years, I'll continue to use 12.75 + NNUE that is still the best NOT Stockfish clone available, excluding Dragon 2.
Just updated my SuperBlitz 41 adding also Ethereal 13.00 github no NNUE 1 CPU (or it crashes on Windows 10|11 ARM64) that is very weak and cannot even compete with low-end older engines like Xiphos 0.6!
Please Andrew, Ethereal was a myth 2 years ago, almost as strong as Stockfish 10!!! Fix crashes and compile it for Windows 10 |11 ARM64, Linux ARM64, Raspberry PI4 and Mac M1 NEON, letting us to test your algorithms also with Stockfish free test NNUEs, Why you limit the github version when all other original and SF derivatives programmers are adding NNUE to their engines with big improvement??
AlexChess wrote: ↑Fri Jul 09, 2021 8:15 am
Why you limit the github version when all other original and SF derivatives programmers are adding NNUE to their engines with big improvement??
Diversity. All the MANY engines reusing the SF toolkit are doing nothing but producing a worse Stockfish.
#WeAreAllDraude #JusticeForDraude #RememberDraude #LeptirBigUltra "Those who can't do, clone instead" - Eduard ( A real life friend, not this forum's Eduard )
AlexChess wrote: ↑Fri Jul 09, 2021 8:15 am
Why you limit the github version when all other original and SF derivatives programmers are adding NNUE to their engines with big improvement??
Diversity. All the MANY engines reusing the SF toolkit are doing nothing but producing a worse Stockfish.
Sorry, I respect your opinion, but I don't agree with you.
MikeB's (sf contributor) Honey, Harmon, Black Diamond, Bluefish, Oki Maguro, Shashchess 17.1 (Andrea Manzo sf contributor) SugaR AI (Marco Zerbinati also sf contributor) Corchess 1.4, but also Fischerov, Killfish, Eman, Zeus and others that do not release source but give their work for free often score MUCH BETTER than Stockfish and they deserve my attention playing very different than it.
AlexChess wrote: ↑Fri Jul 09, 2021 8:15 am
Why you limit the github version when all other original and SF derivatives programmers are adding NNUE to their engines with big improvement??
Diversity. All the MANY engines reusing the SF toolkit are doing nothing but producing a worse Stockfish.
Sorry, I respect your opinion, but I don't agree with you.
MikeB's (sf contributor) Honey, Harmon, Black Diamond, Bluefish, OKI Maguro, Shashchess 17.1 (Andrea Manzo sf contributor) SugaR AI (Marco Zerbinati also sf contributor) Corchess 1.4, but also Fischerov, Killfish, Eman, Zeus and others that do not release source but give their work for free often score MUCH BETTER than Stockfish and they deserve my attention playing very different than it.
I don't believe a single one of those engines is worth your time, and I think you would agree if you knew more about how little they differ from Stockfish in the places where it matters.
#WeAreAllDraude #JusticeForDraude #RememberDraude #LeptirBigUltra "Those who can't do, clone instead" - Eduard ( A real life friend, not this forum's Eduard )
AlexChess wrote: ↑Fri Jul 09, 2021 8:15 am
Why you limit the github version when all other original and SF derivatives programmers are adding NNUE to their engines with big improvement??
Diversity. All the MANY engines reusing the SF toolkit are doing nothing but producing a worse Stockfish.
Sorry, I respect your opinion, but I don't agree with you.
MikeB's (sf contributor) Honey, Harmon, Black Diamond, Bluefish, OKI Maguro, Shashchess 17.1 (Andrea Manzo sf contributor) SugaR AI (Marco Zerbinati also sf contributor) Corchess 1.4, but also Fischerov, Killfish, Eman, Zeus and others that do not release source but give their work for free often score MUCH BETTER than Stockfish and they deserve my attention playing very different than it.
I don't believe a single one of those engines is worth your time, and I think you would agree if you knew more about how little they differ from Stockfish in the places where it matters.
It's called "fine tuning" ...Also Stockfish is derived from Glaurung. it is perfectly allowed and brings new ideas to the Stockfish Team (if the code is open)
AlexChess wrote: ↑Fri Jul 09, 2021 8:15 am
Why you limit the github version when all other original and SF derivatives programmers are adding NNUE to their engines with big improvement??
Diversity. All the MANY engines reusing the SF toolkit are doing nothing but producing a worse Stockfish.
Sorry, I respect your opinion, but I don't agree with you.
MikeB's (sf contributor) Honey, Harmon, Black Diamond, Bluefish, OKI Maguro, Shashchess 17.1 (Andrea Manzo sf contributor) SugaR AI (Marco Zerbinati also sf contributor) Corchess 1.4, but also Fischerov, Killfish, Eman, Zeus and others that do not release source but give their work for free often score MUCH BETTER than Stockfish and they deserve my attention playing very different than it.
I don't believe a single one of those engines is worth your time, and I think you would agree if you knew more about how little they differ from Stockfish in the places where it matters.
It's called "fine tuning" ...Also Stockfish is derived from Glaurung. it is perfectly allowed and brings new ideas to the Stockfish Team (if the code is open)
And as an FYI: You said "Fischerov, Killfish, Eman, Zeus and others that do not release source but give their work for free often score MUCH BETTER than Stockfish". If the authors of those versions have given the binary to anyone, then they are violating Stockfish's GPLv3 and are unlawful. I suggest you don't support people who care so little about the code they take as to not even comply with the basic license. Also you should ask yourself why the authors don't release the "modified" source.
#WeAreAllDraude #JusticeForDraude #RememberDraude #LeptirBigUltra "Those who can't do, clone instead" - Eduard ( A real life friend, not this forum's Eduard )
Yes I agree with you. It's really bad not sharing the SF GPLv3 source code modified to the chess community. I have asked it to all authors of these engines so I could compile them for Linux ARM64, Windows ARM64 and Mac M1 and the WHOLE community would benefit for their innovations. Instead, I'm really grateful to all authors that share it without limitations.
Finally completed the deep Ethereal 13.00 standard vs 12.75 + NNUE match. As expected the first is much weaker. Since Ethereal 13.00 github is 2400 ELO and Ethereal 12.75 is 2404 ELO (according to SPCC, https://www.sp-cc.de/index.htm) I can extrapolate that "my" FrankenstEthereal is already a little stronger than Ethereal 13.00 AVX2 + NNUE commercial, that is rated 2476 ELO, but the quest for the NNUE Graal for Ethereal 12.75 is just started... I will try all the best 21 mb NNUE, not only from SF.
(Sorry, Ethereal 13.00 can use only 1 CPU or it crashes on my system, but I have preferred to use the official Grant's build to the Mac Silicon M1 github build, waiting somewone that could compile the github version also for Windows 11 ARM64, like Igel 3.0.5 and Rubichess 2.2 that are working at full speed )
Good news!
With the right net Ethereal 13.00 commercial could be strongly improved
Ethereal 12.75 sse3 github with Zeus.net is already stronger than Fat Fritz 2.0 (Chessbase, Stockfish 13 FF 2.1 github version is much stronger) and Dragon 2.0.
Please Andrew, unlock NNUE on the free version, to let testers to experiment. If you like I can give you all the games 12.75 NNUE games + Zeus.net (used also by AI-15) to study why with alien nets Ethereal become so strong, and rebuild your own net.
I think that testing 2 versions of the same engine is not enough, the performance must be calculated considering all the best engines availables.
PS: @Stefan Waiting for your SPCC accurate ranking with the "alien nets" with 6000 games, if possible
Now switching from Arena 3.5.1 to Banksiagui 0.46 that is more stable on engine matches.